There has since been an increase in the needs for more intuitionally observing a structure of an object under observation by reconstructing a three-dimensional structure from a two-dimensional image acquired by an electron microscope, and extracting an optional spot from a reconstruction image to be thereby quantified. In order to make such an observation as described, use is generally made of a back-projection method (refer to Nonpatent Document 1, pp. 228-231) whereby respective continuously tilted two-dimensional images obtained by continuously tilting an object under observation are back-projected from respective tilt angles at the time of image acquisition to thereby reconstruct a three-dimensional image, and a technique (refer to Nonpatent Document 1, pp. 231-234) whereby the respective continuously tilted two-dimensional images are assumed as true-value images in respective projection directions, and algebratical reconstruction is executed such that an error between the projection result of each unknown three-dimensional image and a relevant true-value image will be at the minimum.
Further, as techniques for reducing an information-incomplete region, which is a problem point of those techniques described as above, there are available various techniques including a technique (refer to refer to Nonpatent Document 1, pp. 225-228 and p. 264) whereby a sample is tilted at sample-tilt axes, differing from each other, and a plurality of three-dimensional reconstruction images, each thereof using respective tilt-images as photographed, are superimposed one on another, a Dots concentration reconstruction technique (refer to refer to Nonpatent Document 2) whereby an image is regarded as an aggregate of dots, and dots configuration of a three-dimensional image is optimized on the basis of the respective continuously tilted two-dimensional images, and a discrete algebraic reconstruction technique (Dart) (refer to refer to Nonpatent Document 3) whereby a reconstruction image serving as a reference is created by an algebraic reconstruction method, the reconstruction image is divided into regions according to a threshold, and subsequently, the algebraic reconstruction method is applied again using information pieces in the respective regions.